Continuous Learning Models Across Edge Hierarchies

1. A computer-implemented method for training models, comprising: receiving, by a first server from a second server of a plurality of servers, inference data generated based on a first model, wherein content of data captured by one or more sensor devices is evaluated by the second server based on the first model to generate the inference data, the inference data describes identifying an object in the content of the captured data, and the second server of the plurality of servers comprises less computing resources than the first server; determining, by the first server, a data drift in the received inference data generated based on the first model, wherein the data drift indicates a degree of inaccuracy of the inference data that identify the object in the content of the captured data; determining that the data drift exceeds a drift threshold; evaluating computing resources of each of the plurality of servers; determining whether the second server has sufficient computing resources to train a second model while continuing to generate inference data; in response to determining the second server does not have sufficient computing resources, determining at least a third server of the plurality of servers that has sufficient computing resources to train the second model; requesting at least the third server to train the second model; receiving the trained second model from the third server; and causing, by the first server, the first model to be updated with the trained second model on a graphics processing unit (GPU) of the second server of the plurality of servers, thereby causing the second server to execute an inference operation on the GPU using trained second model” and replace with the first model updated with the trained second model for improved accuracy of inferencing a subsequently captured sensor data.

2. The computer-implemented method of claim 1, wherein the first server is one of a network edge server or a cloud server of a multi-access edge computing hierarchy, and wherein the second server and the third server are either an on-premises edge server or a network edge server of the multi-access edge computing hierarchy.

3. The computer-implemented method of claim 2, the method further comprising: determining that the third server has sufficient computing resources for training the second model, wherein the third server is distinct from the second server; requesting the third server to train the second model; and receiving from the third server the trained second model.

4. The computer-implemented method of claim 2, the method further comprising: responsive to determining that the data drift exceeds the drift threshold, searching for the trained second model in a model cache at the first server; and retrieving the second model from the model cache.

5. The computer-implemented method of claim 2, wherein the drift threshold corresponds to a predetermined inference accuracy level, the data drift is caused by a brightness of a scene changing over time affecting depiction of the content of data captured by the one or more sensor devices, and wherein a model is selected for retraining when inferences generated by the model are below the predetermined inference accuracy level.

6. The computer-implemented method of claim 2, wherein the inference data is generated by the first server based on captured data received from one or more devices over a wireless network.

7. The computer-implemented method of claim 2, wherein the inference data includes one or more regions of interest in an image captured by an imaging device.

8. A system for training models, the system comprising: a processor; and a memory storing computer-executable instructions that when executed by the processor cause the system to: receive, by a first server from a second server of a plurality of servers, inference data generated based on a first model, wherein content of data captured by one or more sensor devices is evaluated by the second server based on the first model to generate the inference data, the inference data describes identifying an object in the content of the captured data, and the second server of the plurality of servers comprises less computing resources than the first server; determine, by the first server, a data drift in the received inference data generated based on the first model, wherein the data drift indicates a degree of inaccuracy of the inference data that identify the object in the content of the captured data; determine that the data drift exceeds a drift threshold; evaluate computing resources of each of the plurality of servers; determine whether the second server has sufficient computing resources to train a second model while continuing to generate inference data; in response to determining the second server does not have sufficient computing resources, determine at least a third server of the plurality of servers that has sufficient computing resources to train the second model; request at least the third server to train the second model; receive the trained second model from the third server; and cause, by the first server, the first model to be updated with the trained second model on a graphics processing unit (GPU) of the second server of the plurality of servers, thereby causing the second server to execute an inference operation on the GPU using the first model updated with the trained second model for improved accuracy of inferencing a subsequently captured sensor data.

9. The system of claim 8, wherein the first server is one of a network edge server or a cloud server of a multi-access edge computing hierarchy, and wherein the second server and the third server are either an on-premises edge server or a network edge server of the multi-access edge computing hierarchy.

10. The system of claim 9, the computer-executable instructions when executed further cause the system to: determine that the third server has sufficient computing resources for training the second model, wherein the third server is distinct from the second server; request the third server to train the second model; and receive from the third server the trained second model.

11. The system of claim 9, the computer-executable instructions when executed further cause the system to: responsive to determining that the inference data exceeds the drift threshold, search for the trained second model in a model cache at the first server; and retrieve the second model from the model cache.

12. The system of claim 9, wherein the drift threshold corresponds to a predetermined inference accuracy level, the data drift is caused by a brightness of a scene changing over time affecting depiction of the content of data captured by the one or more sensor devices, and wherein a model is selected for retraining when inferences generated by the model are below the predetermined inference accuracy level.

13. The system of claim 9, wherein the inference data is generated by the first server based on captured data received from one or more devices over a wireless network.

14. The system of claim 9, wherein the inference data includes one or more regions of interest in an image captured by an imaging device.

15. A non-transitory computer-readable recording medium storing computer-executable instructions that when executed by a processor cause a computer system for training models to: receive, by a first server from a second server of a plurality of servers, inference data generated based on a first model, wherein content of data captured by one or more sensor devices is evaluated by the second server based on the first model to generate the inference data, the inference data describes identifying an object in the content of the captured data, and the second server of the plurality of servers comprises less computing resources than the first server; determine, by the first server, a data drift in the received inference data generated based on the first model, wherein the data drift indicates a degree of inaccuracy of the inference data that identify the object in the content of the captured data; determine that the data drift exceeds a drift threshold; evaluate computing resources of each of the plurality of servers; determine whether the second server has sufficient computing resources to train a second model while continuing to generate inference data; in response to determining the second server does not have sufficient computing resources, determine at least a third server of the plurality of servers that has sufficient computing resources to train the second model; request at least the third server to train the second model; receive the trained second model from the third server; and cause, by the first server, the first model to be updated with the trained second model on a graphics processing unit (GPU) of the second server of the plurality of servers, thereby causing the second server to execute an inference operation on the GPU using the first model updated with the trained second model for improved accuracy of inferencing a subsequently captured sensor data.

16. The computer-readable recording medium of claim 15, wherein the first server is one of a network edge server or a cloud server of a multi-access edge computing hierarchy, and wherein the second server and the third server are either an on-premises edge server or a network edge server of the multi-access edge computing hierarchy.

17. The computer-readable recording medium of claim 16, the computer-executable instructions when executed further cause the system to: determine that the third server has sufficient computing resources for training the second model, wherein the third server is distinct from the second server; request the third server to train the second model; and receive from the third server the trained second model.

18. The computer-readable recording medium of claim 16, the computer-executable instructions when executed further cause the system to: responsive to determining that the data drift exceeds the drift threshold, search for the trained second model in a model cache at the first server, wherein the data drift is caused by a change on a scene between a sunny daylight and an overcast over time affecting depiction of the content of data captured by the one or more sensor devices; and retrieve the second model from the model cache.

19. The computer-readable recording medium of claim 16, wherein the inference data is generated by the first server based on captured data received from one or more devices over a wireless network.

20. The computer-readable recording medium of claim 16, wherein the inference data includes one or more regions of interest in an image captured by an imaging device.